AFFECTIVE RESPONSE OF MUSICAL STIMULI PREDICTOR
A NEURAL NETWORK MODEL
PSYC450: Neural Network Models of Social + Cognitive Processes
Dr. Stephen Read | University of Southern California | Fall 2013
Given a musical stimulus, this neural network model, created with Emergent Neural Network Simulator, can predict the elicited affective response of a listener.
HOW IT WORKS
I trained the network to measure the psychoacoustic factors of tempo, volume, sharpness, and novelty of a stimulus, and use the data to quantify a listener's experience of arousal and valence.
Because the relationship between music and emotion is not linear, this model takes a spatiotemporal approach by utilizing attributes of long and short-term memory in the form of connection weights.
INPUT LAYERS -- 10x1 GEOMETRY
1. Tempo /beats per measure
2. Volume /piano to forte
3. Sharpness /staccato to legato
4. Novelty /degree of familiarity
Input layers composed of ten nodes each, creating a sliding scale that can measure continuous quantities.
HIDDEN LAYER -- 25x1 GEOMETRY
Input layers connected to hidden layer; hidden layer bi-directionally connected to output layers.
OUTPUT LAYERS -- SCALAR VALUES
1. Arousal /physiological response
2. Valence /positive or negative emotional value